Really great piece. We have recently seen many popular lemmy instances struggle under recent scraping waves, and that is hardly the first time its happened. I have some firsthand experience with the second part of this article that talks about AI-generated bug reports/vulnerabilities for open source projects.

I help maintain a python library and got a bug report a couple weeks back of a user getting a type-checking issue and a bit of additional information. It didn't strictly follow the bug report template we use, but it was well organized enough, so I spent some time digging into it and came up with no way to reproduce this at all. Thankfully, the lead maintainer was able to spot the report for what it was and just closed it and saved me from further efforts to diagnose the issue (after an hour or two were burned already).

The theory that the lead maintainer had (he is an actual software developer, I just dabble), is that it might be a type of reinforcement learning:

• Get your LLM to create what it thinks are valid bug reports/issues
• Monitor the outcome of those issues (closed immediately, discussion, eventual pull request)
• Use those outcomes to assign how "good" or "bad" that generated issue was
• Use that scoring as a way to feed back into the model to influence it to create more "good" issues

If this is what's happening, then it's essentially offloading your LLM's reinforcement learning scoring to open source maintainers.

Kiriha (the character in the lead image) is a Windows advocate in the story. She is brutal...

My goodness, I hadn't realized how big fedinsfw.app had become. It's now easily the biggest piefed instance with almost 2x the MAU of piefed.social.

I work professionally in the field of pharmaceutical manufacturing, though mainly with liquid and lyophilized presentations. The issue talked about in this video, polymorphism and isomerization, would primarily just be an issue with drugs that are considered small molecules. Historically, that is most drugs...but that is changing as more and more other modalities are hitting the market.

Small molecules, especially in a dry powder, undergo lots of processes that would make cleaning with enough rigor to completely eliminate a polymorph extremely difficult. Things like milling, mixing, spray drying, and pressing are extremely energetic and intentionally disperse powder like crazy. The cleaning cycles for all this equipment are already extremely rigorous, often reducing contaminants to levels too low to measure and then undergoing sterilization cycles with vaporized hydrogen peroxide (VHP) between batches. However, when you are talking about autocatalysis like this, it only takes ppb-level contamination to propagate through the whole batch.

When you are talking about large molecules like biologics (antibodies, most vaccines, etc.), there are a whole bunch of other things that can go wrong, but not this. The most analogous thing I can think of would be a persistent viral contamination. Current good manufacturing practices (cGMP) should be able to alleviate this though, and it wouldn't impact tech transfer to a different site like what happened in the video. Alternatively, I have worked on mAbs where each batch would measure the glycosylation state because it impacted activity. That is a post-translational modification process for that drug that some batches worked better than others, but we never really tracked down a root cause (before the program was killed).

Another case study that happened on a program I was working on was that we had a series of batches where the active ingredient (a mAb) was oxidized at a much higher rate than normal. Nothing about our process had changed, but it was a clear night/day difference that happened all of a sudden. After lots of testing and ruling things out, we managed to figure out that there were elevated levels of iron in our product. By sampling at each step of our process, we figured out that the iron was being introduced during two steps called bioburden reduction filtration and sterile filtration. You might notice that these are both unit operations where our drug passes through filter membranes (they were in fact the same filters). Ultimately, it turned out that the supplier of our filters had a contamination issue that impacted their filters from a certain lot and later. So, we were ultimately able to fix it, but I remember those reports being literally hundreds of pages of ruling stuff out before we finally figured out the root cause.

Hope that helps you feel better!

I agree that it is unlikely to be a chemical reaction. Instead, I suspect you are nucleating bubbles for the dissolved gasses in the solution (think diet coke and mentos).

When yeast is active, it creates CO2 as a byproduct. This is how you get bubbles on the top of your solution when you are blooming the yeast. This gas byproduct is also dissolved into the water as well. Letting it sit for that long would give it plenty of time to completely saturate the liquid (or even supersaturate it depending on environmental changes).

When you throw the salt in, those salt crystals act as a nice nucleation point allowing those dissolved gases to form a bubble and leave the liquid phase. I can't really speak to how salt behaves differently to flour in this regard, but nucleation rate is proportional to the available surface area for nucleation, and salt crystals tend to not clump up nearly as much as flour, making more surface area accessible for nucleation.

I suspect you already know this since you make so much bread, but, I was always taught that introducing an osmotic shock like adding salt directly to your yeast will slow down the rising/proofing process. So, it would be best to add the flour, then the salt on top, then mix them together to help blunt the osmotic stress on your yeast.

For those out there poking the code, please disclose responsibly! Don't just make a public post about a security vulnerability, reach out to the devs first to give them a chance to create a fix.

More people need to make use of the curation tools available to them in my opinion. Too often I see people browsing /all and then complain about seeing tons of stuff that they don't like. There are tools that are available to them to help them create a more pleasant experience that they just aren't using. The fediverse doesn't have some algorithm that learns the type of content that you most often engage with and feed it to you, you have to more proactively do the curation yourself.

Just as an example, I was in a conversation with one of the lemmy.world admins a while back. We were talking about instance blocks and how infrequently users actually use them. Across all of the users on lemmy.world, only about 700 of them actually created an instance block for the most-blocked instance (lemmynsfw). Only two instances had more than 500 users block it.

I mean...that doesn't really seem that bad? Also, asking for the whole community to be nsfw is a wild overreaction looking at the other pictures in the community.

If you browse the all feed, expect to see some things you don't like/enjoy. It's a fire hose of content by design. Learn to curate your subscribed feed and stick to it. Frankly, lemmy doesn't have great filters/blocks to do what you want, and expecting the whole rest of the internet to abide by such strict standards of nsfw isn't going to happen.

You have clearly never driven on 93 through Boston where the person you replied to said they are from (aka the Big Dig). It is basically an entire highway that is underneath the city. There are many on and off ramps, lanes suddenly become exit only, complex multi-lane exits that branch...it's intimidating. As somebody that has lived in the Boston area for 15 years now, I still mess things up.

I realized that I replied from the wrong browser tab. I can confirm that this is not some impersonator wjs018 🕵️

wjs018 [is] now admin

Hey! That's me! I just wanted to provide some thoughts on this release...

Most of my dev time for this release went into backend api stuff. Specifically, this point:

Automatically-updating Swagger docs to ease frontend development

So, most of the regular users probably won't really notice (unless it breaks your favorite mobile app). However, I just wanted to send some thanks to both rimu and freamon for the work that they did in finishing this up. Also, to thank the cadre of mobile app/frontend developers that provided input, tested changes, and pointed out problems along the way. It was a lot of work that is completely transparent to most people, at least it hopefully is transparent.

Going forward, the newly schematized api should be much easier for developers interacting with it, and we shouldn't be introducing breaking changes at this point. I wouldn't yet call the api complete or mature, but it is at least a stable base from which we can expand, and the most common types of interactions should be implemented at this point.

The other feature I wanted to highlight is this one:

Federate flair changes properly

This was actually a lot of work for such a short bullet point. Basically, we rewrote the post flair system so that it federated out as part of the community in the activitypub messages passed between instances. What this means for users is that communities located on piefed 1.2+ instances should have their full flair list visible on remote instances, regardless of whether the flair has been used or not. For example, if a community mod changes the name of a flair, it will automatically be updated on every instance and every post that it was attached to as that flair edit is federated out.

This work also laid the groundwork for post flair being interoperable with what lemmy is planning to do with their 1.0 release. Their work is still ongoing, so it might yet change, but when we were restructuring how flair was implemented on our end, we tried to make it look like what they are currently planning so that it can hopefully be a seamless integration when lemmy 1.0 is released.

Admins: I also added a function for remote communities that lets you "fix up from remote" located near the bottom of the sidebar. Basically, if you have a remote community on your instance that has some extraneous pre-1.2 flair hanging around that could use some cleaning up, then you can hit that button and it should phone back home to the community's home instance and kind of reset things for you to bring everything in sync.

Same...

I will just take this opportunity to highlight some updates to the API endpoints coming in the next version of PieFed (so far, we are actively working on more still):

• Tons more user settings (I basically made it possible to control every single setting using either the web UI or the API)
• Poll post support (including voting)
• Event post support
• Unsubscribe from everything endpoint (warning, this is irreversible)
• Rating communities
• Marking a comment as an answer (aka Stack Overflow)
• The mark as AI label

In general, it has been a priority for us to make sure that when we add features these days, that we also make a corresponding API endpoint to do that same action. There is still plenty of things not in the API, but we are working our way through the backlog.

Lastly, I just want to send out a huge thanks to all the app/frontend developers that have very patiently been helping us debug all these things as we have been implementing them on the test instance. Also, a thanks to users like Sunshine that make sure to keep the pressure on us to make sure we don't forget anything!

Official response from Greg Bernhardt

It's years since I last used PhysicsForums, but found it immensely useful in the old days while going through my undergrad physics degree (it was less useful for PhD courses). I am not morally opposed to providing AI attempts at an answer in threads where nobody else chimes in. However, using real accounts that belong to other users is wildly over the line. I was surprised to see this wasn't really called out in the official response thread by the existing users as that is the part of all this that is the most egregious to me.

Rust vs. python might be part of it since python is basically designed to be quick to write. However, the lemmy devs, to their credit, are extremely thorough. They make sure that any feature they implement is fully available in the UI, the API, and has tests written for it.

For piefed, we test stuff as we work on it...but there isn't a formal test suite or anything like that. And, as evidenced by this post, many new features only show up in the web ui and may get added to the api later, time/effort depending. This is something we have been trying to get better at, adding features to both the web ui and the api at the same time, but it is more work so it goes slower.

This feature helps avoid reaching the block limit.

There is a block limit? I don't think that's right...but I haven't looked too much at the block code other than to make it work without needing a page refresh

I loved this show. It was so needlessly well produced. Probably one of the darkest comedies I have seen in some time.

The manga is great as well.

There is an admin setting that requires a user to be logged in before viewing pages on the site. We keep that off normally, but we recently had it on for a couple of days because of over-aggressive scrapers.

I can definitively say that the only way to access topics and feeds is through the web ui. Either the desktop interface or through the PWA on a mobile device would do it.

The reason no other apps have topic/feed support yet is because topics and feeds haven't yet been added to the piefed api. It's something that we want to do, but the api is still very new for piefed, and it is very much a work in progress, adding features as we get the chance to code them. I do know that it is something that the app developers want too (example). So, it will take some time, but I expect it to come sooner or later.

This is pretty much always going to be the case because the database query is much more complicated to filter by your subscriptions rather than the completely unfiltered views that are just sorted a certain way.

I don't have direct experience working in the lemmy codebase, but I do in piefed, and this is just an inherent limitation of how databases work. Basically, there is a big table in the db with all the posts that the instance knows about. As an example, let's say that you want to look at your Subscribed feed and sort it by New. First, the database sorts all the posts by newest first (this step is very quick). Then it will filter out all the entries that are not in a community that you are subscribed to (this step is a lot slower). So, views where the database doesn't need to do that filtering step are a lot faster than others. There are other things that slow down queries as well like blocks or keyword filters.

Smarter admins than I can employ strategies to tune their database for better performance. I don't have any knowledge in this area though, so others would be able to provide more insightful answers.